907993 - Annotate crash reports if the slow-script dialog has been shown

Status: RESOLVED WONTFIX

(Core :: XPConnect, defect)

Description

[Benjamin Smedberg]

We had a theory that some crashes are much more likely to happen during/after the slow-script dialog. We'd like to annotate crash reports when that dialog is first shown using a key.

Comment 1

[(not currently active) Ted Mielczarek]

What would we want to add here? A boolean if we've ever seen a slow script dialog? A timestamp of when the slow script dialog was last shown?

Comment 2

[Benjamin Smedberg]

For now I just want a boolean.

Comment 3

[Sanchit Garg]

hey Benjamin,
I am interested in solving this bug. Please guide me.

Comment 4

[Benjamin Smedberg]

Sanchit, crash report annotations are added via this interface: http://mxr.mozilla.org/mozilla-central/source/xpcom/system/nsICrashReporter.idl#64http://mxr.mozilla.org/mozilla-central/source/xpcom/system/nsICrashReporter.idl#64

The slow script dialog is shown starting here: http://hg.mozilla.org/mozilla-central/annotate/4887845b1142/dom/base/nsGlobalWindow.cpp#l9320

And you can probably add the annotation at or near the top of the function.

As a bonus, it may be interesting to annotate also whether the user killed the script from the dialog.

I don't know if there are any tests for the slow-script dialog which you can copy to test this annotation; bholley do you know?

Comment 5

[Bobby Holley (:bholley)]

(In reply to Benjamin Smedberg  [:bsmedberg] workweek high latency 19-Aug through 23-Aug from comment #4)
> Sanchit, crash report annotations are added via this interface:
> http://mxr.mozilla.org/mozilla-central/source/xpcom/system/nsICrashReporter.
> idl#64http://mxr.mozilla.org/mozilla-central/source/xpcom/system/
> nsICrashReporter.idl#64
> 
> The slow script dialog is shown starting here:
> http://hg.mozilla.org/mozilla-central/annotate/4887845b1142/dom/base/
> nsGlobalWindow.cpp#l9320
> 
> And you can probably add the annotation at or near the top of the function.
> 
> As a bonus, it may be interesting to annotate also whether the user killed
> the script from the dialog.
> 
> I don't know if there are any tests for the slow-script dialog which you can
> copy to test this annotation; bholley do you know?

I recently wrote tests for the watchdog thread, but not the slow-script dialog itself. Having some would probably be very worthwhile - it'd involve hooking up a mock prompt service, dropping the max script runtime pref, and then causing script to infinite loop.

Comment 6

[Sanchit Garg]

Hey Benjamin,
when can I catch you on irc?
:)

Comment 7

[Benjamin Smedberg]

I'm "bsmedberg", US Eastern Time (UTC-4).

Comment 8

[y.anvesh]

Attachment: Patch to annotate slow script

We have this test file which annotates the slow script but we do not know how to test the annotation is present.

Comment 9

[Benjamin Smedberg]

Testing this is hard because we don't expose the annotations in-process so you'd actually have to crash the process to check them. But an automated test may not be worthwhile in this case.

Comment 10

[Benjamin Smedberg]

Comment on attachment 8376705 [details] [diff] [review]
Patch to annotate slow script

Skip the test and fix the indentation in the #ifdef MOZ_CRASHREPORTER case. Then set r?bholley on the new patch. Thank you!

Comment 11

[vinayakagarwal6996]

Hello,
Is the bug still available to fix?

Thanks

Comment 12

[Josh Matthews [:jdm]]

Is it still worthwhile for someone to address comment 10?

Comment 13

[Benjamin Smedberg]

Hrm, things have changed somewhat with e10s. In e10s mode, we no longer spin a nested event loop for the slow-script dialog, so the crashiness is less interesting or serious in that case. 

So no, this is probably no longer worth fixing.